diff --git a/docs/src/SUMMARY.md b/docs/src/SUMMARY.md
index 057d691..be04873 100644
--- a/docs/src/SUMMARY.md
+++ b/docs/src/SUMMARY.md
@@ -4,6 +4,7 @@
 # User Guide
 
 - [Installing Keyfork](./INSTALL.md)
+- [Security Considerations](./security.md)
 - [Shard Commands](./shard.md)
 - [Common Usage](./usage.md)
 - [Configuration File](./config-file.md)
diff --git a/docs/src/security.md b/docs/src/security.md
new file mode 100644
index 0000000..4256e8a
--- /dev/null
+++ b/docs/src/security.md
@@ -0,0 +1,81 @@
+# Security Considerations
+
+Keyfork handles data that is considered sensitive. As such, there are a few
+base considerations we'd like to make about the environment Keyfork is run in.
+This ensures that the amount of mitigations needed to run Keyfork are reduced.
+
+## Build Process
+
+Keyfork should be built using a secure toolchain, such as through Guix or the
+Distrust Packages system. Using something such as `rustup` means Rust can't
+properly be verified from source. Ideally, Keyfork should be built by multiple
+developers and verified between them to ensure the results are deterministic.
+
+## Hardware
+
+Keyfork is expected to run on hardware detached from the Internet and from any
+other computers. This helps ensure the Keyfork seed is never exposed to any
+online system. Exposing the Keyfork seed may result in a compromise of data
+derived from Keyfork. The hardware is expected to be stored in a safe location
+along with the removable storage containing the operating system and (if using
+Keyfork Shard) the shard file, where adversaries are not able to tamper with
+the hardware, OS, or shard file.
+
+## Software
+
+Keyfork is intended to be one of few programs running on a given system. The
+ideal system to run Keyfork under is an OS whose only dependencies are Keyfork
+and Keyfork's runtime dependencies. Because of these restrictions, Keyfork does
+not necessarily need to include memory-locking or memory-hardening
+functionality, although such functionality may be included upon further
+releases.
+
+## Keys in Memory
+
+As Keyfork is expected to be the only program running on a given system, it is
+not expected for Keyfork to defend against malicious software on a system
+scanning the memory of Keyfork and extracting the keys. As such, at this time,
+Keyfork does not zero out previously-used memory. Additionally, if such
+software did exist, because Keyfork is intended to run on hardware detached
+from the Internet and from any other computers, the risk of practical covert
+channels is reduced. Tempest and side channel attacks may be mitigated by
+running Keyfork on hardware located in a Faraday cage.
+
+## Security of Local Shards
+
+The threat model of Keyfork in a "local shard" configuration is that an
+adversary can, without leaking the seed:
+
+* Compromise `M-1` shard holders or shards
+
+The threat model of Keyfork in a "local shard" configuration does not include:
+
+* Compromise of the system running Keyfork
+
+Keyfork does not provide a mechanism by itself to ensure the operating system
+or the Keyfork binary has not been tampered with. Users of Keyfork on a shared
+system should verify the system has not been tampered with before starting
+Keyfork.
+
+## Security of Remote Shards
+
+The threat model of Keyfork in a "remote shard" configuration is that an
+adversary can, without leaking the seed:
+
+* Compromise `M-1` shard holders, shard holder devices, or shards
+* Eavesdrop upon (but not intercept or tamper with) secure communications
+
+The threat model of Keyfork in a "remote shard" configuration does not include:
+
+* The compromise of the system initiating the "remote shard" requests.
+
+Keyfork has a "remote shard" mode, where shards may be transport-encrypted to
+an ephemeral key and combined on a system run by a user we will call the
+"administrator". In this design, it is expected that a secure communications
+channel is established that can be spied upon but can't be tampered with. The
+administrator can then begin distributing encoded (not encrypted!) public keys
+to remote shardholders, who then decrypt and re-encrypt the shards to an ECDH
+AES-256-GCM key. Because the shard is re-encrypted, it can't be intercepted by
+anyone intercepting the communication. However, it is possible for the
+administrator to leak either the Keyfork seed or any number of shards if they
+are the only user operating the system combining the shares.